vLLM
A fast and easy-to-use library for LLM inference and serving.
News
25.9.2026 vLLM version 0.29.0 installed on Roihu, adding support to many new models
7.4.2026 vLLM now available as a separate module on Roihu-GPU
Available
Currently supported vLLM versions:
| Version | Module | Notes |
|---|---|---|
| 0.29.0 | python-vllm/0.29.0 |
Default |
| 0.19.1 | python-vllm/0.19.1 |
|
| 0.18.0 | python-vllm/0.18.0 |
Includes vLLM, PyTorch and related libraries with GPU support via CUDA/ROCm.
PyTorch vs vLLM
If you don't particularly need vLLM, we recommend using the full PyTorch module instead, which includes more packages.
If you find that some package is missing, you can often install it
yourself using pip install. It is recommended to use Python virtual
environments. See our Python documentation for more information on
how to install packages
yourself.
If you think that some important package should be included in the
module provided by CSC, please contact our
servicedesk.
If adding a few packages with pip is not enough, you can also build
your own containers, for example by extending CSC's machine learning
containers.
All modules are based on containers using Apptainer (previously known
as Singularity). Wrapper scripts have been provided so that common
commands such as python, python3, pip and pip3 should work as
normal.
License
vLLM is covered by the Apache License 2.0.
Usage
To load the default version on Roihu-GPU:
If you wish to have a specific version (see above for available versions), use:
To check the exact packages and versions included in the loaded module you can run:
Example script for usage
See the vLLM section in CSC's Machine learning guide, which has links to example script for using vLLM on Roihu.
The full Machine learning guide might also be of use.